1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
|
dnl AMD64 mpn_mod_34lsub1 -- remainder modulo 2^48-1.
dnl Copyright 2000-2002, 2004, 2005, 2007, 2010-2012 Free Software Foundation,
dnl Inc.
dnl This file is part of the GNU MP Library.
dnl
dnl The GNU MP Library is free software; you can redistribute it and/or modify
dnl it under the terms of either:
dnl
dnl * the GNU Lesser General Public License as published by the Free
dnl Software Foundation; either version 3 of the License, or (at your
dnl option) any later version.
dnl
dnl or
dnl
dnl * the GNU General Public License as published by the Free Software
dnl Foundation; either version 2 of the License, or (at your option) any
dnl later version.
dnl
dnl or both in parallel, as here.
dnl
dnl The GNU MP Library is distributed in the hope that it will be useful, but
dnl WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
dnl or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
dnl for more details.
dnl
dnl You should have received copies of the GNU General Public License and the
dnl GNU Lesser General Public License along with the GNU MP Library. If not,
dnl see https://www.gnu.org/licenses/.
include(`../config.m4')
C cycles/limb
C AMD K8,K9 1.0
C AMD K10 1.12
C Intel P4 3.25
C Intel core2 1.5
C Intel corei 1.5
C Intel atom 2.5
C VIA nano 1.75
C INPUT PARAMETERS
define(`ap', %rdi)
define(`n', %rsi)
C mp_limb_t mpn_mod_34lsub1 (mp_srcptr up, mp_size_t n)
C TODO
C * Review feed-in and wind-down code. In particular, try to avoid adc and
C sbb to placate Pentium4.
C * It seems possible to reach 2.67 c/l by using a cleaner 6-way unrolling,
C without the dual loop exits.
ABI_SUPPORT(DOS64)
ABI_SUPPORT(STD64)
ASM_START()
TEXT
ALIGN(32)
PROLOGUE(mpn_mod_34lsub1)
FUNC_ENTRY(2)
mov $0x0000FFFFFFFFFFFF, %r11
sub $2, %rsi
ja L(gt2)
mov (ap), %rax
nop
jb L(1)
mov 8(ap), %rsi
mov %rax, %rdx
shr $48, %rax C src[0] low
and %r11, %rdx C src[0] high
add %rdx, %rax
mov R32(%rsi), R32(%rdx)
shr $32, %rsi C src[1] high
add %rsi, %rax
shl $16, %rdx C src[1] low
add %rdx, %rax
L(1): FUNC_EXIT()
ret
ALIGN(16)
L(gt2): xor R32(%rax), R32(%rax)
xor R32(%rcx), R32(%rcx)
xor R32(%rdx), R32(%rdx)
xor %r8, %r8
xor %r9, %r9
xor %r10, %r10
L(top): add (ap), %rax
adc $0, %r10
add 8(ap), %rcx
adc $0, %r8
add 16(ap), %rdx
adc $0, %r9
sub $3, %rsi
jng L(end)
add 24(ap), %rax
adc $0, %r10
add 32(ap), %rcx
adc $0, %r8
add 40(ap), %rdx
lea 48(ap), ap
adc $0, %r9
sub $3, %rsi
jg L(top)
add $-24, ap
L(end): add %r9, %rax
adc %r10, %rcx
adc %r8, %rdx
inc %rsi
mov $0x1, R32(%r10)
js L(combine)
mov $0x10000, R32(%r10)
adc 24(ap), %rax
dec %rsi
js L(combine)
adc 32(ap), %rcx
mov $0x100000000, %r10
L(combine):
sbb %rsi, %rsi C carry
mov %rax, %rdi C 0mod3
shr $48, %rax C 0mod3 high
and %r10, %rsi C carry masked
and %r11, %rdi C 0mod3 low
mov R32(%rcx), R32(%r10) C 1mod3
add %rsi, %rax C apply carry
shr $32, %rcx C 1mod3 high
add %rdi, %rax C apply 0mod3 low
movzwl %dx, R32(%rdi) C 2mod3
shl $16, %r10 C 1mod3 low
add %rcx, %rax C apply 1mod3 high
shr $16, %rdx C 2mod3 high
add %r10, %rax C apply 1mod3 low
shl $32, %rdi C 2mod3 low
add %rdx, %rax C apply 2mod3 high
add %rdi, %rax C apply 2mod3 low
FUNC_EXIT()
ret
EPILOGUE()
|
